Professional Documents
Culture Documents
Abstract. We show by explicit examples that in many degrees in a stable range the
homotopy groups of the moduli spaces of Riemannian metrics of positive scalar curvature
on closed smooth manifolds can be non-trivial. This is achieved by further developing and
then applying a family version of the surgery construction of Gromov-Lawson to certain
nonlinear smooth sphere bundles constructed by Hatcher.
1. Introduction
1.1. Motivation. Let M be a closed smooth manifold. In this article we study the topology of the space of metrics of positive scalar curvature Riem+ (M ) and of corresponding
moduli spaces. We abbreviate metric of positive scalar curvature by psc-metric.
It has been known for a long time that there are quite a few obstructions to the existence
of psc-metrics. This starts in dimension 2, where the Gau-Bonnet theorem tells us that
only the sphere and RP 2 admit such a metric. In general the Lichnerowicz formula in
combination with the Atiyah-Singer index theorem implies that if M is a spin manifold
non-zero elements in the homotopy groups of Riem+ (M ). More precisely, he proves this
way that 0 (Riem+ (M n )) is non-trivial for n 1, 0, 1 (mod 8) and 1 (Riem+ (M n )) is
non-trivial for n 1, 0 (mod 8).
Contrasting these positive results, it has been an open problem to decide whether
k (Riem+ (M )) for k > 1 or k (Riem+ (M )/Diff(M )) for k > 0 can be non-trivial. Note
that, by construction, Hitchins elements in k (Riem+ (S n )), k = 0, 1, are mapped to zero
in the moduli space Riem+ (M )/Diff(M ). Some experts even raised the suspicion that the
components of this moduli space are always contractible.
1.2. Moduli spaces of psc-merics. In this paper we will construct many examples of
non-zero elements in higher homotopy groups of moduli spaces of psc-metrics on closed
smooth manifolds M . We denote by Riem(M ) the space of all Riemannian metrics with
the C -topology. The group of diffeomorphisms Diff(M ) acts from the right on the space
Riem(M ) by pull-back: (g, ) 7 (g). The orbit space of this action is the moduli
space of Riemannian metrics on M and written M(M ). The orbit space M+ (M ) of the
restricted Diff(M )-action on the subspace Riem+ (M ) of psc-metrics, the moduli space of
Riemannian metrics of positive scalar curvature on M , is our principal object of interest.
In general the action of the full diffeomorphism group is not free on Riem(M ): For
example, if a finite group G acts effectively on M (i.e. if G occurs as a finite subgroup
of Diff(M )), then any metric on M can be averaged over G, and the resulting metric will
be fixed by G. Therefore we also consider the moduli spaces with observer as proposed by
Akutagawa and Botvinnik [2].
1.1. Definition. Let (M, x0 ) be a connected closed smooth manifold with some basepoint
x0 . Let Diffx0 (M ) be the subgroup of Diff(M ) of those diffeomorphisms which fix x0 and
induce the identity on the tangent space Tx0 M . This is the group of diffeomorphisms which
preserve an observer based at x0 .
1.2. Lemma. If (M, x0 ) is a connected smooth closed manifold with a basepoint x0 then
Diffx0 (M ) acts freely on the space Riem(M ) of Riemannian metrics on M .
Proof. This lemma is well known, compare e.g. [7, Proposition IV.5]. For convenience we
recall the proof. Assume g is a Riemannian metric on M , Diffx0 (M ) and g = g .
This means that the map is an isometry of (M, g). As x0 and Tx0 M are fixed by , so
are all geodesics emenating from x0 (pointwise). Since M is closed and connected, every
point lies on such a geodesic, so is the identity.
In the following we equip Diff(M ) and Diffx0 (M ) with the C -topologies. Let Mx0 (M ) =
Riem(M )/Diffx0 (M ). We call Mx0 (M ) the observer moduli space of Riemannian metrics
on M . Since the space Riem(M ) is contractible and the action of Diffx0 (M ) on Riem(M )
is proper (see [10]), Lemma 1.2 implies that the orbit space Mx0 (M ) is homotopy equivalent to the classifying space BDiffx0 (M ) of the group Diffx0 (M ). In particular one obtains
a Diffx0 (M )-principal fiber bundle
(1.3)
the observer moduli space of psc-metrics. Again we obtain a Diffx0 (M )-principal fiber
bundle
(1.4)
Diffx0 (M ) Riem+ (M ) M+
x0 (M ) .
M+
x0 (M )
?
-
Mx0 (M )
Riem+ (M ) M+
x0 (M ) Mx0 (M ).
The constructions of Hitchin [17] use certain non-zero elements in k Diff(M ) and push
them forward to the space Riem+ (M ) via the first map in (1.4). It is then shown that
these elements are non-zero in k Riem+ (M ) (for k = 0, 1).
Our main method will be similar, but starting from the fiber sequence (1.7). We will show
that certain non-zero elements of k BDiffx0 (M ) = k Mx0 (M ) can be lifted to M+
x0 (M ).
Once such lifts have been constructed, it is immediate that they represent non-zero elements
in k M+
x0 (M ) as their images are non-zero in k Mx0 (M ).
1.3. The results. We start from the particular manifold M = S n . Let x0 S n be a base
point. Then the group Diffx0 (S n ) is homotopy equivalent to the group Diff(Dn , Dn ) of
diffeomorphisms of the disk Dn which restrict to the identity on the boundary Dn . These
groups and their classifying spaces have been studied extensively. In particular the rational
homotopy groups q BDiffx0 (S n ) Q are known from algebraic K -theory computations
and Waldhausen K -theory in a stable range.
Theorem 1.8 is essentially an existence theorem and does not directly lead to a geometric
interpretation of the generators of k BDiffx0 (S n ) Q. This was achieved later in the work
of Bokstedt [5] and Igusa [18, 20] based on a construction of certain smooth nonlinear disk
and sphere bundles over S k due to Hatcher. The nontriviality of some of these bundles is
detected by the non-vanishing of a higher Franz-Reidemeister torsion invariant.
Recall from [18, 19, 20] that for any closed smooth manifold M there are universal higher
Franz-Reidemeister torsion classes 2q H 4q (B Torr(M ); Q), where Torr(M ) Diff(M )
is the subgroup of diffeomorphisms of M that act trivially on H (M ; Q). Note that
Diffx0 (S n ) Torr(S n ) and that Torr(S n ) is the subgroup of Diff(S n ) consisting of orientation preserving diffeomorphisms. In particular, these classes define characteristic classes
for smooth fiber bundles M E B over path connected closed smooth manifolds B
with 1 (B) acting trivially on H (M ; Q). (The last condition can be weakend to H (M ; Q)
being a unipotent 1 (M )-module [20], but this is not needed here).
The relevant class 2q H 4q (S 4q ; Q) of the Hatcher bundles over S 4q with fiber S n
was computed in [13, 18, 20] and shown to be non-zero, if n is odd. The generators of
k BDiffx0 (S n ) appearing in Theorem 1.8 can be represented by classifying maps S k
BDiffx0 (S n ) of these Hatcher bundles in this way. In order to prove Theorem 1.9 we
construct families of psc-metrics on these bundles.
Therefore, in Section 2, we will first study how and under which conditions such constructions can be carried out. Assuming that a given smooth bundle admits a fiberwise
Morse function, we use the surgery technique developed by Walsh [30], which generalizes
the Gromov-Lawson construction of psc-metrics via handle decompositions [12, 14], to families of Morse functions, in order to construct families of psc-metrics on this bundle, see
Theorem 2.12. This is the technical heart of the paper at hand. Compared to [30] the
novel point is the generalization of the relevant steps of this construction to nontrivial fiber
bundles.
(1.10)
?
BDiffx0 (S n )
n
This shows that a lift of the class k BDiffx0 (S n ) to k M+
x0 (S ) is nothing but a family
of psc-metrics on the bundle E S k from (1.10).
We will explain the precise relationship in Section 3 and show that the construction
described in Section 2 applies to Hatchers S n -bundles. Here we make use of a familiy of
Morse functions on these bundles as described by Goette [13, Section 5.b]. This will finish
the proof of Theorem 1.9.
Given a closed smooth manifold M of dimension n, we can take the fiberwise connected
sum of the trivial bundle S k M S k and Hatchers exotic S n -bundle. Using additivity
of higher torsion invariants [20, Section 3] we obtain non-trivial elements in k Mx0 (M ) for
given k for any manifold M of odd dimension n as long as k n.
If in addition M admits a psc-metric, this can be combined with the fiberwise psc-metric
on Hatchers S n -bundle constructed earlier to obtain a fiberwise psc-metric on the resulting
nontrivial M -bundle over S k . This shows:
1.11. Theorem. Let M be a closed smooth manifold admitting a metric h of positive scalar
curvature. If dim M is odd, then the homotopy groups k (M+
x0 (M ), [h]) are non-trivial for
0 < k = 4q dim M .
In order to study the homotopy type of the classical moduli space of psc-metrics it remains
to construct examples of manifolds M for which the non-zero elements in k M+
x0 (M ) constructed in Theorem 1.11 are not mapped to zero under the canonical map k M+
x0 (M )
+
k M (M ). This will be done in Section 4 and leads to a proof of the following conclusive
result.
1.12. Theorem. For any d > 0 there exists a closed smooth manifold M admitting a
metric h of positive scalar curvature so that 4q (M+ (M ), [h]) is non-trivial for 0 < q d.
1.13. Remark. One should mention that the manifolds we construct in Theorem 1.12 do
not admit spin structures and are of odd dimension. In particular, the usual methods to
distinguish elements of 0 M+ (M ), which use the index of the Dirac operator, do not apply
to these manifolds, and we have no non-trivial lower bound on the number of components
of M+ (M ).
1.14. Remark. Finding non-zero elements of k Riem+ (M ) for k > 1 remains an open
problem. It would be especially interesting to find examples with non-zero image in
k (Riem+ (M )/Diff(M )), or at least in k (Riem+ (M )/Diffx0 (M )).
We expect that a solution of this problem requires a different method than the one
employed in Sections 3 and 4 of our paper.
1.4. Acknowledgement. Boris Botvinnik would like to thank K. Igusa and D. Burghelea
for inspiring discussions on topological and analytical torsion and thank SFB-478 (Geometrische Strukturen in der Mathematik, M
unster, Germany) and IHES for financial support and hospitality. Mark Walsh also would like to thank SFB-478 for financial support
and hospitality. Thomas Schick was partially supported by the Courant Research Center
Higher order structures in Mathematics within the German initiative of excellence.
2. The surgery method in twisted families
The aim of this section is to prove a result on the construction of fiberwise pcs-metrics
on certain smooth fiber bundles. At first we briefly review the Gromov-Lawson surgery
technique [14] on a single manifold. Here we use the approach developed by Walsh [30, 31].
2.1. Review of the surgery technique on a single manifold. Let W be a compact
manifold with non-empty boundary W and with dim W = n + 1. We assume that the
boundary W is the disjoint union of two manifolds 0 W and 1 W both of which come
with collars
(2.1)
0 W [0, ) W,
1 W (1 , 1] W,
where is taken with respect to some fixed reference metric m on W , see Definition 2.2
below. By a Morse function on W we mean a Morse function f : W [0, 1] such that
f 1 (0) = 0 W,
f 1 (1) = 1 W
and the restriction of f to the collars (2.1) coincides with the projection onto the second
factor
0 W [0, ) [0, ), 1 W (1 , 1] (1 , 1].
We denote by Cr(f ) the set of critical points of f .
We say that a Morse function f : W [0, 1] is admissible if all its critical points have
indices at most (n2) (where dim W = n+1). We note that the last condition is equivalent
to the codimension at least three requirement for the Gromov-Lawson surgery method.
We denote by Morse(W ) and Morseadm (W ) the spaces of Morse functions and admissible
Morse functions, respectively, which we equip with the C -topologies.
2.2. Definition. Let f Morseadm (W ). A Riemannian metric m on W is compatible with
the Morse function f if for every critical point p Cr(f ) with ind p = the positive
M1
q+1
K+
(w)
U
w
W
p+1
K
(w)
M0
p
S
(w)
g1 + dt2
f = c1
transition
transition
g0 + dt2
g0 + dt2
standard
transition
transition
f = c0
f = c0
f = c1
g1 + dt2
Without loss of generality assume that f (w) = 12 . Let c0 and c1 be constants satisfying
0 < c0 < 12 < c1 < 1. The level sets f = c0 and f = c1 divide U into three regions:
U0 = f 1 ([0, c1 ]) U,
Uw = f 1 ([c0 , c1 ]) U,
U1 = f 1 ([c1 , 1]) U.
The region U0 is diffeomorphic to N [0, c1 ]. We use again the flow to identify U0 with
N [0, c1 ] in a way compatible with the identification of W \ U with M0 \ N I . Then,
on U0 , we define g as the product g0 |N + dt2 . Moreover, we extend this metric g0 |N + dt2
near the S p S q I part of the boundary, where again t is the trajectory coordinate.
We will now define a family of particularly useful psc-metrics on the disk Dk . For a
detailed discussion see [30].
2.4. Definition. Let > 0 and be a smooth function : (0, ) R satisfying the
following conditions:
(1) (t) = sin ( t ) when t is near 0;
(2) (t) = when t 2 ;
(3) (t) 0.
Clearly such functions exists, futhermore, the space of functions satifying (1), (2), (3)
for some > 0 is convex. Let r be the standard radial distance function on Rk , and ds2k1
be the standard metric on S k1 (of radius one). Then the metric dr2 + (r)2 ds2k1 on
(0, ) S k1 is well-defned on Rk . By restricting this metric to (0, b] S k1 , one obtains
k
the metric gtor
() on Dk . This metric is defined to be a torpedo metric, see Fig. 3.
10
The most delicate part of the construction, carried out carefully in [30], involves the
following: Inside the region Uw , which is identified with the product Dp+1 Dq+1 , the
p+1
q+1
metric smoothly passes into a standard product gtor
() + gtor
() for some appropriately
chosen , > 0, globally keeping the scalar curvature positive. This is done so that the
induced metric on the level set f 1 (c1 ), denoted g1 , is precisely the metric obtained by
applying the Gromov-Lawson construction to g0 . Furthermore, near f 1 (c1 ) we have
g = g1 + dt2 . Finally, on U1 , which is identified with Dp+1 S q [c1 , 1] in the usual
manner, the metric g is simply the product g1 + dt2 . See Fig. 2 for an illustration.
After the choice of the Morse coordinate diffeomorphism with Dp+1 Dq+1 (and of the
other parameters like and ), the construction is explicit and depends continuously on
the given metric g0 on S p Dq+1 .
Later on we will need the following additional facts. The next lemma is proved in [30,
Section 3].
2.6. Lemma. The initial transition consists of an isotopy. In particular, g0 is isotopic
to a metric which, on a neighborhood diffeomorphic to S p Dq+1 of the surgery sphere Sp
q+1
().
in M0 , is 2 ds2p + gtor
2.7. Lemma. The whole construction is O(p + 1) O(q + 1)-equivariant.
Proof. By construction, the standard product of torpedo metrics even is O(p+1)O(q+1)invariant. It is a matter of carefully going through the construction of the transition metric
in [30] to check that this construction is equivariant for the obvious action of these groups.
This is done in [31, Lemma 2.2].
Lemma 2.7 will be of crucial importance later, when in a non-trivial family we cannot
choose globally defined Morse coordinates giving diffeomorphisms to Dp+1 Dq+1 (as the
bundle near the critical set is not trivial). We will construct Morse coordinates well defined
up to composition with elements of O(p+1)O(q+1). The equivariance of Lemma 2.7 then
implies that our construction, which a priori depends on the choice of these coordinates, is
consistent and gives rise to a smooth globally defined family of metrics.
We should emphasize that this construction can be carried out for a tubular neighbourhood N of arbitrarily small radius and for c0 and c1 chosen arbitrarily close to 21 . Thus
the region Uw , on which the metric g is not simply a product and is undergoing some kind
of transition, can be made arbitrarily small with respect to the background metric m. As
critical points of a Morse function are isolated, it follows that this construction generalizes
easily to Morse functions with more than one critical point.
2.2. Extension to families. There is a number of ways to generalize the surgery procedure to families of manifolds. A construction relevant to our goals leads to families of
Morse functions, or maps with fold singularities. We start with a local description.
11
Rk Rn+1 Rk R,
(y, x) 7 y, c x21 x2 + x2+1 + + x2n+1 .
p2
Rk Rn+1 Rk R R
with the projection p2 onto the second factor defines a Rk -parametrized family of Morse
functions of index on Rn+1 in standard form.
Let W be a compact manifold with boundary W 6= , dim W = n + 1. We denote by
Diff(W, W ) the group of all diffeomorphisms of W which restrict to the identity near the
boundary W . Then we consider a smooth fiber bundle : E B with fiber W , where
dim B = k and dim E = n + 1 + k . The structure group of this bundle is assumed to be
Diff(W, W ) and the base space B to be a compact smooth manifold. Assume that the
boundary W is split into a disjoint union: W = 0 W t 1 W .
Let 0 : E0 B , 1 : E1 B be the restriction of the fiber bundle : E B to the
fibers 0 W and 1 W respectively. Since each element of the structure group Diff(W, W )
restricts to the identity near the boundary, the fiber bundles 0 : E0 B , 1 : E1 B
are trivialized:
0
1
E0 = B 0 W
B, E1 = B 1 W
B.
Choose a splitting of the tangent bundle E of the total space as E
= B Vert , where
Vert is the bundle tangent to the fibers W , i.e. choose a connection.
2.10. Definition. Let : E B be a smooth bundle as above. For each z in B let
iz : Wz E
be the inclusion of the fiber Wz := 1 (z). Let F : E B I be a smooth map. The map
F is said to be an admissible family of Morse functions or admissible with fold singularities
with respect to if it satisfies the following conditions:
(1) The diagram
F BI
E
p
?
1
B
commutes. Here p1 : B I B is projection on the first factor.
(2) The pre-images F 1 (B {0}) and F 1 (B {1}) coincide with the submanifolds
E0 and E1 respectively.
(3) The set Cr(F ) E of critical points of F is contained in E\(E0 E1 ) and near each
p1
critical point of F the bundle is equivalent to the trivial bundle Rk Rn+1 Rk
so that with respect to these coordinates on E and on B the map F is a standard
map Rk Rn+1 Rk R with a fold singularity as in Definition 2.8
12
fz = F |Wz : Wz {z} I I
is an admissible Morse function as in Subsection 2.1. In particular, its critical points
have indices n 2.
We assume in addition that the smooth bundle : E B is a Riemannian submersion
: (E, mE ) (B, mB ), see [4]. Here we denote by mE and mB the metrics on E and
B corresponding to the submersion . Now let F : E B I be an admissible map
with fold singularities with respect to as in Definition 2.10. If the restriction mz of the
submersion metric mE to each fiber Wz , z B , is compatible with the Morse function
fz = F |Wz , we say that the metric mE is compatible with the map F .
2.11. Proposition. Let : E B be a smooth bundle as above and F : E B I be
an admissible map with fold singularities with respect to . Then the bundle : E B
admits the structure of a Riemannian submersion : (E, mE ) (B, mB ) such that the
metric mE is compatible with the map F : E B I .
Proof. One can choose a Riemannian metric mB on the base B , and for each fiber Wz
there is a metric mz compatible with the Morse function fz = F |Wz . Using convexity of
the set of compatible metrics and the local triviality in the definition of a family of Morse
functions, we can choose this family to depend continuously on z . Then one can choose
an integrable distribution (sometimes called connection) to construct a submersion metric
mE which is compatible with the map F : E B I , see [4].
Below we assume that the fiber bundle : E B is given the structure of a Riemannian
submersion : (E, mE ) (B, mB ) such that the metric mE is compatible with the map
F : E B I.
Consider the critical set Cr(F ) E . It follows from the definitions that Cr(F ) is
a smooth k -dimensional submanifold in E , and it splits into a disjoint union of path
components (folds)
Cr(F ) = 1 t t s .
Furthemore, it follows that the restriction of the fiber projection
|j : j B
is a local diffeomorphism for each j = 1, . . . , s. In particular, |j is a covering map, and
if the base B is simply-connected then |j is a diffeomorphism onto its image.
Since the metric mE is a submersion metric, the structure group of the vector bundle
Vert E is reduced to O(n + 1). Futhermore, since the metrics mz are compatible with
the Morse functions fz = F |Wz , the restriction Vert|j to a fold j Cr(F ) splits further
orthogonally into the positive and negative eigenspaces of the Hessian of F . Thus the
metric mE induces the splitting of the vector bundle
Vert|
= Vert Vert +
j
13
with structure group O(p + 1) O(q + 1) for each j . Here is the main result of this
section:
2.12. Theorem. Let : E B be a smooth bundle, where the fiber W is a compact
manifold with boundary W = M0 t M1 , the structure group is Diff(W, W ) and the base
space B is a compact smooth simply connected manifold. Let F : E B I be an
admissible map with fold singularities with respect to . In addition, we assume that the
fiber bundle : E B is given the structure of a Riemannian submersion : (E, mE )
(B, mB ) such that the metric mE is compatible with the map F : E B I . Finally, we
assume that we are given a smooth map g0 : B Riem+ (M0 ).
Then there exists a Riemannian metric g = g(g0 , F, mE ) on E such that for each z B
the restriction g(z) = g|Wz to the fiber Wz = 1 (z) satisfies the following properies:
(1) g(z) extends g0 (z);
(2) g(z) is a product metric g (z) + dt2 near M Wz , = 0, 1;
(3) g(z) has positive scalar curvature on Wz .
Proof. We assume that B is path-connected. Let dim B = k , dim W = n + 1. We denote,
as above, Cr(F ) = 1 t t s , where the j is a path-connected fold. For a given point
z B , we denote by fz = F |Wz : Wz I the corresponding admissible Morse function.
The metric g will be constructed by a method which is quite similar to that employed
in the proof of Theorem 2.3. We begin by equipping the boundary component E0 with
the given Riemannian metric g0 . We choose a gradient-like vector field V and use the
trajectory flow of V to extend g0 as a product metric away from the folds Cr(F ). Near
the folds Cr(F ), some modification is necessary. However, roughly speaking, the entire
construction goes through in such a way that the restriction to any fiber is the construction
of Theorem 2.3.
We will initially assume that Cr(F ) has exactly one path-connected component . The
more general case will follow from this by iterated application of the construction. We will
denote by c the critical value associated with the fold , i.e. p2 F () = c I . Let c > 0
be small. Let V denote the normalized gradient vector field associated to F and mE which
is well-defined away from the singularities of F . As F has no other critical values, we use
V to specify a diffeomorphism
0 : E0 [0, c c ] F 1 (B [0, c c ])
(w, t) 7 (hw (t)),
where hw is the integral curve of V beginning at w . In particular, p2 F 0 is the
projection onto [0, c c ]. As the bundle 0 : E0 B is trivial, this gives rise to a
diffeomorphism
B M0 [0, c c ]
= F 1 (B [0, c c ]).
Let gcc denote the metric obtained on F 1 (B [0, c c ]) by pulling back, via this
diffeomorphism, the warped product metric mB + g0 + dt2 . In order to extend this metric
past the fold , we must adapt our construction.
14
Our next goal is to construct a metric gc+c on F 1 (B [0, c + c ]), so that on each fiber
1 (y) F 1 (B [0, c + c ])
the induced metric has positive scalar curvature and is a product near the boundary. Fiberwise, this is precisely the situation dealt with in Theorem 2.3. However, performing this
over a family of Morse critical points, we must ensure compatibility of our construction
over the entire family. The main problem is that our construction depends on the choise
of Morse coordinates, i.e. the diffeomorphism of a neighborhood of the critical point to
Dp+1 Dq+1 . Because of the non-triviality of the bundle, a global choice of this kind is in
general not possible. We will normalize the situation in such a way that we choose diffeomorphisms up to precomposition with elements of O(p + 1) O(q + 1) (in some sense a
suitable reduction of the structure group). We then use Lemma 2.7, that the construction
employed is equivariant for this smaller group O(p + 1) O(q + 1).
Our strategy actually is to use the fiberwise exponential map for mz at the critical set
as Morse coordinates. Because of the canonical splitting Vert| = Vert Vert + with
structure group O(p + 1) O(q + 1) this gives coordinates which are well defined up to
an action of O(p + 1) O(q + 1) (the choice of orthonormal bases in Vert + and Vert ).
However, these coordinates are not Morse coordinates for F . That the metrics mz are
compatible with the Morse function fz only means that this is the case infinitesimally. We
will therefore deform the given Morse function F to a new Morse function F1 for which
our coordinates are Morse coordinates.
We denote by DVert the corresponding disk bundle of radius with respect to the
background metric mE . For each w , we denote by Dw (Vert ) the fiber of this bundle.
If is sufficiently small, the fiberwise exponential map (and local orthonormal bases for
Vert + and Vert ) define coordinates Dp+1 Dq+1 for neighborhoods of the critical point
in each fiber. We use the exponential map to pull back all structures to Dp+1 Dq+1 and,
abusing notation, denote them in the old way. In particular, the function F is defined on
Dp+1 Dq+1 .
Let and r denote the distance to the origin in Dp+1 and Dq+1 , respectively. Then 2
and r2 are smooth functions on the image under the fiberwise exponential map of D(Vert ).
Moreover, define Fstd : DVert R by Fstd := c 2 + r2 . The compatibility condition on
p
3
F and the Taylor expansion theorem imply that F Fstd = O( r2 + 2 ), i.e. F Fstd is
cubic in the mz -distance to the origin.
Choose a sufficiently small > 0 and a smooth cutoff function : R [0, 1] with
(1) (s) = 1 for s <
(2) (s) = 0 for s > 2
(3) |0 (s)| 10/ s R.
p
Then Ft := Fstd + (1 t ( 2 + r2 ))(F Fstd ) provides a homotopy between F = F0
and F1 of families of Morse functions with the following properties:
(1) Cr(Ft ) = Cr(F ) t [0, 1];
(2) Ft coincides with F outside of a tubular neigbourhood of t [0, 1];
15
q+1
Dw
DVert
w
p+1
Dw
Figure 4. The images of the trajectory disks Dwp+1 and Dwq+1 in Dw Vert()
after application of the inverse exponential map
The second and the third condition are evident. For the first, we have to check that we
did not introduce new critical points. Now the gradient of Fstd is easily calculated and its
norm at x is equal to the norm of x. On the other hand, the gradient of
p
(1 t( r2 + 2 ))(F Fstd )
has two summands:
p
(1) (1t( r2 + 2 ))(F F 0 ) where (1t) is bounded and (F F 0 ) is quadratic
in the distance to the origin (as F F 0 has a Taylor expansion which starts with
cubic
pterms).
p
(2) t0 ( r2 + 2 )( r2 + 2 )(F F 0 ). This vanishes
identically if r2 + 2 2 , and
p
is bounded by 10(F F 0 )/0 10(F F 0 )/ p
r2 + 2 if r2 + 2 2 (here we use
that the gradient
of the distance to the origin r2 + 2 has norm 1). Since F F 0
p
is cubic in r2 + 2 , the whole expression is quadratic.
It follows that, if is chosen small enough (there is a uniform bound because we deal
with a compact family, so we find uniform bounds for the implicit constants in the above
estimates), the gradient of Ft vanishes exactly at the origin. Near the origin, by the choice
of , Ft = tFstd + (1 t)F . Because the Hessians of F and of Fstd are identical at the
origin, the Hessian of Ft also coincides with the Hessian of Fstd , in particular Ft is a family
16
of Morse functions. To find the required local Morse coordinates, we can invoke Igusas
[18, Theorem 1.4].
Thus we can assume that the map F is standard near the fold , i.e. F = F1 in the first
place, and from now on we will do so. Now, via the fiberwise exponential map for m(w) ,
q+1
DVertw+ = Dw
p+1
DVertw = Dw
Figure 5. The shaded region denotes the region of the fiber Dw Vert(F ) on
which the induced metric is defined.
for each w we can specify a neighbourhood Uw 1 ((w)) containing the point w
and of the type described in the proof of Theorem 2.3. In Figures 4 and 5, the image of
this region under the inverse exponential map, before and after the above adjustment of
F , is shown. For each w , replace the fiber Dw Vert with the image under the inverse
exponential map of Uw . Abusing notation we will retain the name DVert for this bundle,
the fibers of which should be thought of as the cross-shaped region described in Fig. 5.
The structure group of this bundle is still O(p+1)O(q+1). The metric induced by gcc
is defined on a subbundle with fibers diffeomorphic to S p Dq+1 I , see Figure 5. On each
fiber we now perform the construction from Theorem 2.3. The fact that we adjusted F to
make the trajectories standard on the fiber disk guarantees consistency of the construction.
On each fiber there is a splitting into positive and negative eigenspaces over which we
will perform our construction. We must however, choose a pair of orthonormal bases for
the negative and positive eigenspaces of that fiber in order to appropriately identify the
fiber with Euclidean space. In order to guarantee consistency we must ensure that our
construction is independent of these choices. But this follows from Lemma 2.7.
Extending the metric fiberwise in the manner of Theorem 2.3 and pulling back via the
exponential map, gives a smooth family of fiber metrics, which, with respect to some
17
integrable distribution H and the base metric mB , combine to the desired submersion
metric on F 1 (B [0, c + c ]).
2.13. Remark. With some little extra care it should be possible to remove the condition
that B is simply connected in Theorem 2.12. However, we are only interested in the case
B = S n with n > 1 so that, for our purpose, we can stick to the simpler version as stated.
3. Metrics of positive scalar curvature on Hatchers examples
The work of Goette [13, Section 5.b] shows that Hatchers examples can be given the
structure which is described in Definition 2.10. The construction of the Hatcher bundles
Dn E S k is explained in some detail in [13] and will not be repeated here. Most
important for our discussion is the fact that each of these bundles comes with an admissible
family F of Morse functions as indicated in the following commutative diagram:
Dzn
iz
(3.1)
F =(,f )
S k [0, 1/2]
?
p1
Sk
We follow the description given in [13]. Each fz := f |Ez : Dzn [0, 1/2] has three critical
(0)
(1)
(2)
(0)
points pz , pz and pz . In particular, the points pz form a unique fiberwise minimum
of the Morse functions fz with value 0, and F 1 (S k {0}) has a neighborhood F 1 (S k
[0, 1/8]) which (as a smooth bundle) is diffeomorphic to Dn S k . Near the value 1/2, the
inverse image F 1 (S k {1/2}) has a neighborhood diffeomorphic to (S n1 I) S k . We
now consider the upside-down copy of the bundle (3.1):
Dzn
iz
(3.2)
F -
S k [1/2, 1]
?
p1
Sk
Here E := E and F (e) := ((e), 1f(e) (e)), i.e. fz = 1fz , where we write F = (, f ).
(0)
(1)
(2)
It follows that each fz : Dn [1/2, 1] has three critical points pz , pz and pz . In
(0)
particular, the points pz form a unique fiberwise maximum of the Morse functions fz
with value 1, and (F )1 (S k {0}) has a neighborhood (F )1(S k [7/8, 1]) which (as a
smooth bundle) is again diffeomorphic to Dn S k . Near the value 1/2, the inverse image
(F )1 (S k {1/2}) again has a neighborhood diffeomorphic to (S n1 I) S k .
By cutting out the neighborhood Dn S k of the fiberwise minima of F , we obtain a
smooth bundle
(S n1 [1/8, 1/2])z
(3.3)
iz
E1
F1
?
p1
Sk
S k [1/8, 1/2]
18
19
deform the metric h near x0 such that its restriction on D0n is a torpedo metric gtor (r)
(use e.g. Lemma 2.6 and thinking of the given disk as one half of a tubular neighborhood
of an embedded S 0 ). Thus we will assume that the metric h already has this property.
On the other hand, we consider the bundle E S k with with psc-metrics gz on the
fibers Szn , z S k , as constructed before. We notice that the metrics gz are chosen in such
way that their restrictions to the disks (D0n )z and (D1n )z are torpedo metrics (with chosen
parameter). Let
fn z = S n \ (Dn )z .
D
z
1
This is a disk together with the metric gez = g z |D
which is a product-metric g0 + dt2 near
n
g
z
fn z . Now for each z S k we define the Riemannian manifold
the boundary Szn1 D
fn z
Mz = M #(S n )z = (M \ D0 ) Szn1 D
equipped with the metric e
hz so that
e
hz |M \D0 = h|M \D0 ,
e
hz |Dzn = gez .
20
/ Riem(M )/G
Riem(M )G EDiff(M ) = BG
/ Riem(M )/ Torr(M ) o
Riem+ (M )/ Torr(M )
Riem+ (M )/Diff(M )
/ Riem(M )/Diff(M ) o
21
collapses at the E2 -level. From this it follows that the map induces an isomorphism in
sheaf theoretic cohomology with rational coefficients.
In order to derive the statement of the lemma, note that up to homotopy equivalence
the space BDiff(M ) can be assumed to be a paracompact Frechet manifold [22, Section
44.21], in particular to be locally contractible. This and the homotopy invariance of sheaf
theoretic cohomology [8, Theorem II.11.12] imply by [8, Theorem III.1.1.] that there is a
canonical isomorphism
H (BDiff(M ); Q)
= H (BDiff(M ); Q)
sh
sing
22
4.4. Lemma. For each n 3, there is an orientable closed hyperbolic (hence aspherical)
n-manifold B n so that Out(1 (B n )) = 1.
Next, let k 2 be a natural number. We construct an orientable 4k -dimensional
manifold N as follows.
Recall the Moore space M (Z/2, 2) = S 2 D3 where : D3 S 2 is of degree 2. Its
reduced integral homology is concentrated in degree 2 and isomorphic to Z/2. Let S 2
B SO(3k) represent a generator of 2 (B SO(3k)) = Z/2. This map can be extended to a
map M (Z/2, 2) B SO(3k) which then induces an isomorphism H 2 (B SO(3k); Z/2)
=
H 2 (M (Z/2, 2); Z/2) of groups that are isomorphic to Z/2. By pulling back the universal bundle over B SO(3k) we obtain a Euclidean vector bundle X M (Z/2, 2) of
rank 3k which is orientable, but not spin. At this point we note that the generator
of H 2 (B SO(3k); Z/2) is the second Stiefel-Whitney class of the universal bundle over
B SO(3k).
In this discussion we can replace M (Z/2, 2) by a homotopy equivalent finite 3-dimensional
simplicial complex, which we denote by the same symbol. If k is chosen large enough then
M (Z/2, 2) can be embedded as a subcomplex into Rk+1 . We consider a regular neighbourhood R Rk+1 of this subcomplex. This is an compact oriented submanifold of Rk+1 with
boundary which contains M (Z/2, 2) as a deformation retract. By construction R is an
oriented closed smooth manifold of dimension k . Furthermore, because R has the rational
homology of a point, Poincare duality and the long exact homology sequence for the pair
(R, R) show that R is a rational homology sphere. Let E R be the restriction of
the pull back over R of the vector bundle X M (Z/2, 2). If k is chosen large enough,
then H 2 (R; Z/2) H 2 (R; Z/2) is an isomorphism and hence E is not spin.
Let DE be the disc bundle of E and let P be the oriented double of DE . The manifold
P is the total space of an oriented S 3k bundle over R with vanishing Euler class (the
latter for dimension reasons). Hence the rational homology of P is concentrated in degrees
0, k , 3k and 4k and isomorphic to Q in these degrees. Furthermore, the manifold P is
orientable, but not spin. The latter holds, because the tanget bundle of DE restricted
to R splits as a direct sum T (R) E and the bundle T (R) is stably trivial, since it
becomes trivial after adding a trivial real bundle of rank 1.
Because P is simply connected by construction, the Hurewicz theorem modulo the Serre
class of finite abelian groups shows that P has finite homotopy groups up to degree k 1.
If we additionally assume that k is odd, then the only possibly non-zero Pontrijagin class
A-genus
of Q is nonzero.
23
For later use we recall that positive definite lattices have finite automorphism groups:
Given such a lattice E choose a bounded ball D around 0 which contains a set of generators.
Because E is finitely generated and positive definite, D is finite. Now observe that each
automorphism E permutes the points in D and is uniquely determined by this permutation.
We finally define the oriented manifold N 4k := P ]Q as the connected sum of P and Q.
4.5. Lemma. For each odd n > 0 and each (sufficiently large and odd) k > n, the manifold
M := B n N 4k has all the properties described in Proposition 4.1.
Proof. The dimension of M is odd and can be chosen arbitrarily large.
The manifold N is simply connected, of dimension at least 5 (if k is large enough) and
not spin. It therefore carries a metric of positive scalar curvature [14] and the same is then
true for the product B n N 4k .
Because B is aspherical and N is simply connected, we can regard the projection p1 :
M = B N B onto the first factor as the classifying map of the universal cover
of M . By construction, the manifold M has finite 2 and 4 and the higher A-genus
n
n
hA(M ) (c), [M ]i associated to the fundamental class c H (B; Q) = H (B1 (M ); Q)
is nonzero. Because the group 1 (B) is the fundamental group of a hyperbolic manifold, it
is torsion free and does not contain Z2 as a subgroup (the latter by Preissmans theorem).
This implies that the image of any homomorphism Z center(1 (M )) is trivial. We
can therefore apply [16, Theorem 4.1.] to conclude that M does not carry any effective
S 1 -action.
We now show that AH Aut(H (M ; Q)) is finite. Let f : M M be a diffeomorphism.
Up to homotopy we can assume that f fixes a base point x0 so that we get an induced
automorphism f : 1 (M, x0 ) 1 (M, x0 ) and together with the classifying map p1 : M
B a homotopy commutative diagram
/
M
B
Bf
24
f; Q).
lattice is positive definite, the map f can induce only finitely many self maps of H 2k (M
f is concentrated in degrees 0, k , 3k and
The remaining nonzero rational cohomology of M
4k where it is isomorphic to Q. Hence f can act only by minus or plus the identity on
these cohomology groups.
We conclude that f preserves the subspaces p1 (H (B n ; Q)) and p2 (H (N ; Q)) of the
vector space H (M ; Q) and can only act as the identity on the first and in finitely many
ways on the second. Because H (M ; Q) is generated as a ring by these subspaces, f is
determined by the action on these subspaces. This shows that AH is indeed finite.
f; Q) must
The preceding argument also shows that the induced action of f on H 4k (M
be the identity, since a generator of this group can be chosen as the k -th Pontrijagin class
f by the construction of Q. This and the fact that f acts trivially on H n (M ; Q) =
of M
H n (B n ; Q) (see above) imply that f must act in an orientation preserving fashion on the
manifold M .
References
[1] M. Belolipetsky, A. Lubotzky, Finite groups and hyperbolic manifolds, Invent. Math. 162 (2005),
no. 3, 459472.
[2] K. Akutagawa, B. Botvinnik, The Relative Yamabe Invariant, Comm. Anal. Geom. 10 (2002) 925-954.
[3] K. Akutagawa, B. Botvinnik, Manifolds of positive scalar curvature and conformal cobordism theory,
Math. Ann. 324 (2002) 817-840.
[4] A. Besse, Einstein manifolds. Ergebnisse der Mathematik und ihrer Grenzgebiete, Springer-Verlag,
Berlin, 1987.
[5] M. B
okstedt, The rational homotopy type of WhDiff () , Algebraic topology, Aarhus 1982 (Aarhus,
1982), 2537, Lecture Notes in Math., 1051, Springer, Berlin, 1984.
[6] B. Botvinnik, P. Gilkey, The eta invariant and metrics of positive scalar curvature. Math. Ann. 302
(1995), no. 3, 507517.
[7] J.-P. Bourguignon, Une stratification de lespace des structues riemanniennes, Comp. Math. 30 (1)
(1975), 1-41.
[8] G. E. Bredon, Sheaf theory, Graduate texts in Mathematics 170, Springer-Verlag 1997.
[9] V. Chernysh, On the homotopy type of the space R+ (M ) , math.GT/0405235.
[10] D. Ebin, The manifold of Riemannian metrics. 1970 Global Analysis (Proc. Sympos. Pure Math.,
Vol. XV, Berkeley, Calif., 1968) pp. 1140 Amer. Math. Soc., Providence, R.I.
[11] F. T. Farrell, W. C. Hsiang, On the rational homotopy groups of the diffeomorphism groups of discs,
spheres and aspherical manifolds. Algebraic and geometric topology (Proc. Sympos. Pure Math.,
Stanford Univ., Stanford, Calif., 1976), Part 1, pp. 325337, Proc. Sympos. Pure Math., XXXII,
Amer. Math. Soc., Providence, R.I., 1978.
[12] P. Gajer, Riemannian metrics of positive scalar curvature on compact manifolds with boundary.
Ann. Global Anal. Geom. 5 (1987), no. 3, 179191.
[13] S. Goette, Morse theory and higher torsion invariants I, math.DG/0111222.
[14] M. Gromov, H. B. Lawson, The classification of simply connected manifolds of positive scalar curvature. Ann. of Math. (2) 111 (1980), no. 3, 423434.
[15] M. Gromov, H. B. Lawson, Positive scalar curvature and the Dirac operator on complete Riemannian
25